Model Improvement Support System Data Confidentiality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a challenge in developing and improving machine learning models while maintaining data confidentiality and restricting access to sensitive datasets, as the indicators and datasets focused on by model developers may differ from those prioritized by application developers.
Innovation Solution
A model improvement support system and method that allow model developers to select datasets and execute learning/evaluation programs while ensuring that the execution conditions satisfy the dataset's associated conditions, thereby enabling model improvement without disclosing sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the application developer provides all datasets to the model developer for model improvement, then the model improvement can be thoroughly achieved, but the data confidentiality and security are compromised
Solution Approach 1:
The patent introduces a learning/evaluation program as an intermediary that operates on data without requiring the model developer to directly access or view the actual data content. The program acts as a mediator between the application developer's data and the model developer's improvement needs, enabling model training while maintaining data confidentiality through controlled data processing environments.
Solution Approach 2:
The patent creates a virtual copy of the data processing environment where the learning/evaluation program can replicate the necessary data processing functions without transferring actual sensitive data. This allows the model developer to work with data representations that maintain confidentiality while still enabling thorough model improvement through the copied processing logic.
2Reliability
If the model developer accesses the dataset directly to improve the model, then the model can be improved using actual data, but the access control and security restrictions are violated
Solution Approach 1:
The learning/evaluation program is designed to be self-contained with built-in execution condition verification. It automatically checks whether its own execution conditions are satisfied by the provided data before proceeding, eliminating the need for complex external access control mechanisms while still ensuring secure data processing for model improvement.
Solution Approach 2:
The patent performs preliminary verification of execution conditions against the provided data before allowing any data processing to occur. This advance checking prevents unauthorized or inappropriate data access from the outset, simplifying the overall access control architecture by addressing security concerns before they can manifest as complex control mechanisms.
3Object-affected harmful factors
If the execution conditions are strictly verified against dataset attributes, then the data confidentiality is maintained, but the model improvement efficiency is reduced
Solution Approach 1:
The patent merges the execution condition verification functionality directly into the learning/evaluation program itself, combining the data processing tasks with the security verification tasks. This integration allows both confidentiality maintenance and improvement efficiency to be achieved simultaneously, as the same program that processes data for model improvement also verifies execution conditions against dataset attributes.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The model improvement support system makes a determination, for each of one or more datasets selected by a model developer from among one or more datasets provided from an application developer and input to the model in utilization of the model, of whether or not the execution condition of the learning/evaluation program for performing learning/evaluation on the model satisfies an execution condition associated with the dataset, wherein the learning/evaluation is at least either of learning and evaluation of the model, and executes the learning/evaluation program with this dataset used as an input to the model if the result of the determination is affirmative.